Wan 2.2 I2V for Low VRAM (GGUF)

My personal modification of HazardAI's Wan I2V workflow, modified to use quantized GGUF models instead of FP16 checkpoints so that it can run on low VRAM.

Models

Unet

The following files should be saved into /models/unet

• Wan 2.2 I2V A14B HighNoise Q4_K_M GGUF

• Wan 2.2 I2V A14B LowNoise Q4_K_M GGUF

Text Encoders

The following files should be saved into /models/text_encoders

• NSFW Wan UMT5-XXL fp8-scaled

VAE

The following files should be saved into /models/vae

• Wan 2.1 VAE

LoRAs (for speed)

The following files should be saved into /models/loras

• Wan 2.2 I2V lightx2v 4-steps LoRA v1 High Noise

• Wan 2.2 I2V lightx2v 4-steps LoRA v1 Low Noise